The impact of shade tree species identity on coffee pests and diseases

The multifunctional role of shade trees for conservation of biodiversity and ecosystem services in natural forests and agroforests is well documented, yet we lack insights into the impact of shade tree species identity on pest and disease dynamics on agroforestry crops such as coffee and cacao


Introduction
Trees play a major role in the functioning of a broad range of ecosystems, such as forests, agroforests and semi-open landscapes. Within these ecosystems, they influence processes related to soil fertility, erosion, water quality and carbon storage (Beer et al., 1998;Brockerhoff et al., 2017;Tscharntke et al., 2011). Trees also influence the microclimate by providing shade, which in turn affects organismal physiology and the diversity and composition of the understory plant community Lin, 2007;Schroth et al., 2000). But not all tree species are alike. Trees differ in the quantity and quality of light that penetrates their canopy, whether or not they associate with beneficial organisms like nitrogen-fixing bacteria, and the identity of pests, pathogens and natural enemies that associate with their canopy and roots (Buba, 2015;Staver et al., 2001). As such, tree species identity, density and spatial distribution can have a major impact on the diversity and composition of the understory plant community (Babin et al., 2010;Bedimo et al., 2008;Gidoin et al., 2014;Schroth et al., 2000). Tree species identity may also influence the interactions between understory plants and other organisms, such as herbivores and pathogens (Chamagne et al., 2016;De Groote et al., 2017;Staver et al., 2001), which is something we lack insight into. Such information is important, as it can contribute to our fundamental understanding of spatial variation in pest and pathogen dynamics on understory plants. From an applied perspective, the informed selection and planting of trees has the potential to reduce pest and disease levels, and might therefore contribute to an increase in crop yield in agroforestry systems and the conservation of rare understory plants in natural systems (Barrios et al., 2012;Durand-Bessart et al., 2020;Gras et al., 2016).
Agroforestry, where agricultural crops such as coffee and cacao are cultivated under a canopy of shade trees, is often considered to be a sustainable type of land use (Perfecto et al., 2014). Shade trees frequently provide habitat for the natural enemies of pests and pathogens, and thereby enhance pest and disease control (Vandermeer et al., 2010). Yet, it is commonly assumed that shade trees are equal in their ability to regulate pests and pathogens (Narango et al., 2019). In many traditional agroforestry systems, crops were cultivated under the dense, multilayered canopy of a diverse set of native shade trees. However, in a push to increase yields, farmers have reduced shade levels, and during this process retained only selected shade trees species (Geeraert et al., 2019a). This selection process is based on farmer's knowledge of ecosystem services and disservices provided by each tree species (Ango et al., 2014). These management decisions often leave a strong imprint on the composition of shade trees in agroforests, and this might have profound effects on pest and disease levels on crops. Yet, farmers may overlook the impact of tree identity on pest and disease levels, and instead select trees based on characteristics such as rapid growth to quickly fill a gap in the canopy, crown architecture, leaf characteristics (size and shape) and association with nitrogen fixing bacteria, with the aim to increase plant health, growth, and soil fertility (Beer et al., 1998;Tscharntke et al., 2011). Moreover, they might select certain shade trees for other benefits, such as timber and suitability of the flowers for honey bees (Ango et al., 2014;Lamond et al., 2019;Muleta et al., 2011). Understanding the relationship between shade tree species identity and pest and disease levels in agroforestry crops like coffee could complement the list of selection criteria used by farmers, and thereby contribute to the natural suppression of pest and disease levels.
There are several mechanisms by which shade tree identity might impact pests and pathogens in agroforestry systems. Trees differ in their canopy architectural characteristics and leaf traits, resulting in differences in the quality and quantity of light, temperature, humidity and amount of rainwater underneath the canopy, thereby altering the understory microclimate (Gagliardi et al., 2021;Lemenih et al., 2004;Mouen Bedimo et al., 2008, 2007Sercu et al., 2017). Differences in canopy characteristics, leaf traits, leaf and root chemistry, as well as the association of the tree with nitrogen-fixing bacteria, might affect the physical and chemical characteristics of the soil (Beer et al., 1998;Muleta et al., 2008;Staver et al., 2001). These changes in the microclimate and soil fertility may shape the diversity and composition of several organisms that live underneath the canopy, including pests, pathogens and their natural enemies Boudrot et al., 2016;Gagliardi et al., 2021;Teodoro et al., 2008). Canopy functional traits, such as crown shape and leaf type, as well as nitrogen fixing ability of shade trees, are often considered as selection criteria by farmers in agroforestry systems, because of their perceived influence on yield and pest and disease incidence (Albertin and Nair, 2004;Cerdán et al., 2012;Valencia et al., 2015). Likewise, canopy cover, which determines light interception, is often considered as it might affect pest and disease incidence by modifying the microclimate Gagliardi et al., 2021;Staver et al., 2001). Shade trees may compete with coffee shrubs for soil moisture, especially during the dry season (Beer et al., 1998), and this may cause drought stress and increased susceptibility to pests and diseases (Durand-Bessart et al., 2020;Schroth et al., 2000). Another mechanism by which shade tree species identity may impact the level of pests and diseases is when the shade tree species themselves harbor the pests and pathogens that attack the crop growing underneath them. For example, Room and Smith (1975) reported that the use of Leucaena leucocephala as a shade tree in cacao agroforests was problematic since it had high densities of the cacao armyworm, Tiracola plagiata, which also attacks cacao plants.
Arabica Coffee (Coffea arabica L.) is native to the moist Afromontane forests of southwestern Ethiopia, where it grows naturally as an understory shrub under a species rich canopy of native shade trees (Schmitt et al., 2010). Within the same landscape, Arabica coffee is produced along a broad gradient of management intensity, which ranges from coffee shrubs growing under a diverse and dense shade canopy with minimal management through smallholder farms with moderate management to commercial plantation systems with intensive management. At the more intensively managed sites, the diversity and density of shade trees is much reduced, with a dominance of e.g. Acacia abyssinica, Albizia spp., Croton macrostachyus and Cordia africana. The majority of coffee within this landscape is produced by smallholder farmers without the use of pesticides and fertilizer (Zewdie et al., 2020).
In this study, we investigated the impact of shade tree species identity and canopy cover on several major insect pests (coffee blotch miner Leucoptera caffeina, serpentine leaf miner Cryphiomystis aletreuta, coffee leaf skeletonizer Leucoplema dohertyi, antestia shield bugs Antestiopsis spp., and damage by other free-feeding herbivores), diseases (coffee leaf rust Hemileia vastatrix, coffee berry disease Colletotrichum kahawae) and the hyperparasite of coffee leaf rust, Lecanicillium lecanii ( Fig. 1). More specifically, we addressed the following questions: • Do levels of insect pests, diseases and the hyperparasite vary in relation to shade tree species identity, canopy cover and management intensity? • Are soil moisture and shade tree characteristics (nitrogen fixing ability, leaf type and crown shape) driving the effect of shade tree species identity and canopy cover on coffee pests, diseases and the hyperparasite? • What are farmers' decision criteria for selection of coffee shade tree species? Do they consider the relationship between shade tree species identity and pests and diseases?

Study system
The study was conducted in Gomma and Gera districts (7 • 37´-7 • 56Ń and 36 • 13´-36 • 39´E) in Jimma Zone, southwestern Ethiopia (Fig. 1a). Mean annual rainfall in the area is between 1480 and 2150 mm, with the highest amount of rainfall from June to mid-September and mean daily minimum and maximum temperatures are 12 • C and 28 • C, respectively. The landscape was originally covered by moist Afromontane forests, but is currently a mosaic landscape with larger forest remnants, small forest patches, and open areas with annual crop production and communal grazing lands. The study area is one of the centers of origin of Coffea arabica, and coffee is produced along a broad gradient of management intensity (Schmitt et al., 2010;Zewdie et al., 2021). This gradient includes coffee growing with minimal management (i.e., natural forest stand, dense canopy cover, no shade tree pruning, no soil amendments), moderate management (i.e., selective removal of some shade trees, medium level of canopy cover, no soil amendments), and intensive management (i.e., low diversity of shade tree species, low canopy cover, and fertilizer amendments) (Hundera et al., 2013;Tadesse et al., 2014).
The major coffee pests in the study area are the coffee blotch miner Leucoptera caffeina Washbourn, serpentine leaf miner Cryphiomystis aletreuta Meyrick, coffee leaf skeletonizer Leucoplema dohertyi Warren, antestia shield bugs from the genus Antestiopsis, and damage by other free-feeding insects (Mendesil, 2019;Samnegård et al., 2014). The larvae of the coffee blotch miner and serpentine leaf miner produce roundish and serpent-like leaf mines, respectively, while feeding in the upper leaf side (Fig. 1b, c). The coffee leaf skeletonizer feeds on the underside of the leaf, and its damage is easy to recognize as it leaves behind the veins and the upper epidermis (Fig. 1d). Coffee leaves are also fed upon by a diversity of free-feeding herbivores (Fig. 1e), such as the stinging caterpillar Parasa lepida (Cramer) and coffee hawkmoth Cephonodes hylas (Linnaeus). The antestia bug feeds mainly on green berries, but also flower buds and green twigs (Fig. 1f). One of the major fungal pathogens is the coffee leaf rust Hemileia vastatrix Berk and Broome (Fig. 1g). The disease causes defoliation and in the worst case death of branches and heavy crop loss (Avelino et al., 2004). In the study area, the hyperparasitic fungus Lecanicillium lecanii (Zimm.) Zare & W. Gams is often found on the coffee leaf rust lesions (Zewdie et al., 2021). Another major disease is coffee berry disease, which is caused by the pathogen Colletotrichum kahawae (Fig. 1h). The pathogen infects young, developing berries and is characterized by black sunken lesions that lead to mummification of the berries and premature berry drop (Motisi et al., 2019;Waller et al., 2007). For more details on the life-history of the pests and diseases included in this study, see text S1.

Plantation study and landscape survey
To investigate the impact of shade tree species identity on coffee pests, diseases and the hyperparasite, we conducted a detailed empirical study at Horizon coffee plantation in February 2020 and conducted a large-scale survey along a broad gradient of coffee management across the landscape (Fig. 1a). The advantage of the plantation study is thatin stark contrast with the landscape surveythe environment and management are relatively homogeneous. Within the plantation, the shade tree species are growing intermixed and at relatively low density, which makes it easier to identify the influence of shade tree species on coffee pests and diseases and avoids potential biases related to pest and disease aggregation (Cilas et al., 1998) and confounding shade tree species identity with location (Queen et al., 2002). However, patterns uncovered in a relatively homogeneous environment may not reflect those in the more heterogeneous environment of smallholder farms and natural forests (Milcu et al., 2018). For this reason, we complemented the plantation study with a landscape survey of 58 coffee sites in the Gomma and Gera landscape (Fig. 1a).
In the plantation study, we focused on the four most common tree species, namely Acacia abyssinica, Albizia schimperiana, Croton macrostachyus and Cordia africana. We selected ten individuals of each shade tree species, taking care that they were interspersed (i.e. not clustered) in space, and with a minimum distance of 30 m between each of the 40 shade tree individuals. For each shade tree individual, we selected six coffee shrubs underneath the canopy of the respective shade tree, but with variation in the amount of canopy cover, and with a minimum distance of 2 m between the coffee shrubs. In total, we selected 240 coffee shrubs: 4 shade tree species × 10 shade tree individuals per shade The fifty-eight sites (gray circles) and Horizon plantation (red circle) are plotted on the map of the study area. The beige and green background colors represent open and forested areas, respectively. The photos show characteristic damage by the three major coffee insect pests, other free-feeding herbivores, coffee leaf rust, the coffee leaf rust hyperparasite and coffee berry disease: b) a leaf mine of the coffee blotch miner Leucoptera caffeina, c) a leaf mine of the serpentine leaf miner Cryphiomystis aletreuta, d) the coffee leaf skeletonizer Leucoplema dohertyi, e) damage by other free-feeding herbivores, f) antestia shield bug (Antestiopsis spp.), g) coffee leaf rust Hemileia vastatrix (orange spores) and its hyperparasite Lecanicillium lecanii (white spores), and h) mummified coffee berries damaged by coffee berry disease Colletotrichum kahawae.

Table 1
The selection criteria and preferences for shade tree species by coffee farmers in southwestern Ethiopia (N = 58).

Rank
Selection criterion What is preferred by farmers Frequency of response (%) Trees mentioned by farmers tree species × 6 replicate coffee shrubs per shade tree individual.
In the landscape survey, we selected 58 study sites in the Gomma and Gera landscape (Fig. 1a), with the aim to represent the broad gradient of management types characteristic of coffee production in the landscape, ranging from minimal management in the natural forest to moderate management in the smallholder farms and intensive management in commercial plantations (for more details on site selection, see Zewdie et al., 2020). Eight of these sites were within commercial plantations, and similar to the plantation study sites. Within each site, we had previously established a 50 × 50 m plot and labeled 16 coffee shrubs at the intersections of the central 10 m gridlines. For the current study, we randomly selected six of these coffee shrubs. For each coffee shrub, we identified the closest shade tree species. In Table S1, we provide a detailed description of the leaf and canopy characteristics of the shade tree species included in this study. Shade trees were also classified as having feathery leaves (i.e. compound leaves with small leaflets) or non-feathery leaves, as well as crown shape (i.e. widely spreading or non-widely spreading crown; Table S1). We chose these categories because we could directly link them to the wordings used by smallholder farmers in the interviews, in contrast to botanical terms like 'compound' or 'bipinnately' (cf. Table 1). All shade tree species were native to the area.

Recording of pests, diseases and a hyperparasite
As pests have their highest abundance in different seasons (Nestel et al., 1994;Teodoro et al., 2009), we recorded the coffee blotch miner in the wet season of 2018, and the serpentine leaf miner, coffee leaf skeletonizer, free-feeding herbivory and antestia in the dry season of 2019. To record the insect pests, we randomly selected one branch from the lower part of the shrub, and one branch from the upper part of the shrub. We counted the total number of leaves on these two branches, as well as the number of leaves damaged, separately for the coffee blotch miner, serpentine leaf miner, coffee leaf skeletonizer and damage by other free-feeding herbivores ('herbivory'). At the shrub level, we further visually estimated the presence-absence of antestia.
As disease levels can be affected by biennial variation in yield and variation in weather conditions among years (Avelino et al., 2004;Zewdie et al., 2021), we averaged rust and hyperparasite infection levels across the years 2017-2020, and coffee berry disease incidence was averaged across the years 2017-2019. Since the rust is more severe during the dry season as compared to the wet season (Avelino et al., 2018;Zewdie et al., 2021), we surveyed the coffee leaf rust and the fungal hyperparasite during the dry season (February-March), whereas coffee berry disease was assessed during the wet season (July to August). To determine the incidence of coffee leaf rust and the hyperparasite, we recorded the proportion of leaves with rust, and the proportion of rust-infected leaves with the hyperparasite, respectively, on three randomly selected branches (see Zewdie et al., 2021). We further visually estimated coffee leaf rust severity (the percentage of leaf area covered by rust pustules for the subset of leaves with rust infection) and hyperparasite severity (the percentage of rust infected area covered by the hyperparasite) at the shrub level after the survey of the > 60 leaves.
To determine the incidence of coffee berry disease, we recorded the proportion of infected berries on three randomly selected bearing branches (see Zewdie et al., 2020). Plantations in the study area use locally-selected varieties resistant to coffee berry disease, and we thus recorded coffee berry disease only in the smallholder farms (n = 50).

Environmental and management variables
For each individual coffee shrub, we estimated total canopy cover and soil moisture. For both studies, canopy cover was estimated from photos taken from above each coffee shrub in the dry season of 2020 and analyzed using ImageJ v. 1.5.2 (Schneider et al., 2012). Canopy cover ranged from 11% to 93% and 6 91% in the plantation and landscape study, respectively. For the landscape survey, soil moisture was recorded for four consecutive years (2017-2020) during the dry season (January to February), and for the plantation study, soil moisture was recorded in February 2020. We recorded soil moisture in the dry season because shade tree competition is more common and rainfall does not bias the recordings across sites (Beer et al., 1998;Schroth et al., 2000). Average soil moisture content for each shrub was calculated from three recordings at 5 cm depth, taken 10 cm away from the coffee shrub in three directions, using a portable soil moisture probe (SM150, Delta-T Devices Ltd., UK). To characterize management intensity in the landscape survey, we used an index of management intensity that was previously developed for the same set of sites (see Zewdie et al., 2020). This management intensity index characterizes the physical characteristics of the coffee shrubs, and ranges from 1 for a coffee shrub with little or no management to 3 for an intensively managed coffee shrub (Zewdie et al., 2020).

Shade tree preference by farmers
To understand the selection process of shade trees by farmers, and link the observed relationship between shade tree species identity and pests and diseases on coffee to the knowledge of coffee farmers, we interviewed the owners and managers of the 58 study sites (i.e., 50 smallholder farmers and 8 commercial coffee plantation managers). To avoid leading questions, we asked two open-ended questions during a larger interview on coffee diseases, namely 'What shade tree species do you prefer, and which ones do you dislike, for coffee production?' and 'What are the reasons for these preferences and dislikes?'. Interviews were conducted by two persons (BA and Dinkissa Becha).

Statistical analyses 2.6.1. Plantation study
To assess the effect of shade tree species identity and canopy cover on the major insect pests (coffee blotch miner, serpentine leaf miner, coffee leaf skeletonizer), damage by free-feeding herbivores, coffee leaf rust and the hyperparasite of coffee leaf rust, we modeled the proportion of infested leaves out of the total number of leaves assessed on each coffee shrub as a function of shade tree species (Acacia abyssinica, Albizia schimperiana, Croton macrostachyus and Cordia africana) and canopy cover. As species may respond in a non-linear fashion to canopy cover (e. g. if performance is optimal at intermediate shade levels), we further included the quadratic term of canopy cover. We also tested for an interaction between shade tree species identity and canopy cover; as this interactive effect was non-significant for all response variables, we excluded the interaction term from the final model. To account for nonindependence of the six coffee shrubs associated with each shade tree (e. g. due to soil heterogeneity, aggregative patterns of pests and diseases, variation in canopy shape among individuals of the same shade tree species, or differences in the composition of shade trees in the surroundings), we included shade tree individual as a random effect. For the proportional response variables, we assumed a binomial distribution with a logit link. If needed, we accounted for model overdispersion by including coffee shrub identity as a random effect. We conducted similar models for rust severity and hyperparasite severity, where we assumed a Gaussian distribution with identity link, and for the presence of antestia on the shrub, where we assumed a binary distribution and logit link. Statistical outputs of the species-specific models were not corrected for multiple testing, as the high number of species-specific models would have led to strong corrections, possibly obscuring any real effects. Consequently, these results should be interpreted with care, without making strong inferences based on a significant relationship for a single species.

Landscape survey
To assess the effect of shade tree species identity, canopy cover, and management intensity on the insect pests and free-feeding herbivory, we conducted models comparable to those for the plantation study, but with a larger diversity of shade tree species (Acacia abyssinica, Albizia gummifera, Albizia schimperiana, Cordia africana, Croton macrostachyus, Ficus sur, Ficus vasta, Millettia ferruginea, Olea welwitschii, Polyscias fulva, Sapium ellipticum, Schefflera abyssinica, Syzygium guineense and Teclea nobilis) and including the explanatory variable management intensity.
To account for environmental variation, aggregation of pests and diseases and variation in shade tree species among sites, we included 'site' as a random effect. We conducted similar models for rust and hyperparasite incidence and severity (as averaged across four years) and coffee berry disease incidence (as averaged across three years) where we assumed a Gaussian distribution with identity link, and for the presence of antestia on the shrub, where we assumed a binary distribution and logit link.

Soil moisture
To investigate whether soil moisture mediated the effect of shade tree species and canopy cover on the incidence and severity of insect pests and diseases, we ran similar models to those described above, but then we included soil moisture as a covariate. If significant effects of shade tree identity and canopy cover would disappear after including soil moisture as a covariate, we would conclude that soil moisture mediated the effect of shade tree identity and canopy cover on pests and diseases. If the effect of soil moisture was significant, but this did not affect the significance of the effect of shade tree identity and canopy cover, we would conclude that there was an independent effect of soil moisture on pests and diseases.

Leaf type and crown shape
To test for a difference in pest and disease levels on coffee shrubs underneath shade tree species with different characteristics, we estimated three contrasts: i) nitrogen-fixing vs. non-nitrogen-fixing trees, ii) trees with feathery leaves (i.e. compound leaves with small leaflets) vs. trees with non-feathery leaves and iii) trees with a widely spread crown vs. trees with a compact crown. For the classification of shade trees, see Table S1. Contrasts were implemented with the function emmeans in the package emmeans (Lenth et al., 2021).

Shade tree preference by farmers
The preferred shade tree species were tabulated (n = 14 shade trees, Fig. S1), and the reasons for the preference and dislike of shade trees were grouped into 17 selection criteria (see Table 1). Data were then analyzed using frequencies.
All analyses were conducted in R v. 3.6.1 (R Core Team, 2020). We fitted (generalized) linear mixed models with the functions glmer and lmer in the R package lme4 (Bates et al., 2015), evaluated model fit using the packages sjPlot and DHARMa (Hartig and Lohse, 2020;Lüdecke et al., 2020), and tested for significance using the function Anova in the car package (Fox et al., 2020). We scaled continuous variables to mean zero and unit variance (Schielzeth, 2010). All analyses were conducted at the shrub-level. For a detailed overview of statistical models, transformations of response variables, model distributions and link functions, see Tables S2 and S3.

Plantation study
The incidence of insect pests (coffee blotch miner, serpentine leaf miner, coffee leaf skeletonizer, free-feeding herbivory and antestia) did not differ among shade tree species, whereas they often displayed a clear response to canopy cover, but in different ways (Figs. S2 and 2a-e, Table 2). Incidence of the coffee leaf skeletonizer and free-feeding herbivory increased with canopy cover, while the serpentine leaf miner and antestia decreased with canopy cover, and the coffee blotch miner was unaffected by canopy cover (Fig. 2a-e, Table 2).
In contrast to the insect pests, incidence and severity of coffee leaf rust and its hyperparasite significantly differed among shade tree species (Fig. 3, Table 2), but this difference could not be explained by nitrogen fixing ability, leaf type or crown shape of the shade trees (P > 0.05 for all three contrast statements). Rust infection was relatively high underneath the canopy of the shade tree species A. abyssinica (40.8%) and C. macrostachyus (35.1%), and the incidence and severity of the hyperparasite was particularly high on coffee shrubs underneath the canopy of the shade tree A. abyssinica (3.8% and 12.5% for incidence and severity, respectively; Fig. 3). Coffee leaf rust incidence showed a humpbacked relationship with canopy cover, and coffee leaf rust severity linearly decreased with canopy cover (Fig. 2f, g). In contrast to the rust disease, the incidence and severity of the hyperparasite were unrelated to canopy cover (Fig. 2h, i, Table 2).
When including soil moisture as a covariate in the pest and disease models, the effect of shade tree species and canopy cover remained unchanged, suggesting that soil moisture did not mediate the effect of shade tree identity and canopy cover on Arabica coffee pests and diseases (Table S4). We detected an independent, negative effect of soil moisture on free-feeding herbivory, whereas other pests and diseases were unaffected by soil moisture (Table S4 and Fig. S3).

Landscape survey
In contrast to the plantation study, the incidence of the coffee blotch and serpentine miners significantly differed among shade tree species in the landscape survey (Fig. 4, Table 3), but this difference could not be explained by nitrogen fixing ability, leaf type or crown shape of the shade trees (P > 0.05 for all three contrast statements). The coffee blotch miner and antestia decreased with canopy cover, the serpentine leaf miner showed a humpbacked relationship with canopy cover, and the incidence of the coffee leaf skeletonizer and free-feeding herbivory were unrelated to canopy cover (Fig. 5). We did not find a significant effect of management intensity on the pests and diseases (Table 3).
Consistent with the plantation study, coffee leaf rust and its hyperparasite differed significantly among shade tree species, and this difference could not be explained by nitrogen fixing ability and crown shape of the shade trees (Fig. 4, Table 3). However, rust severity was lower on coffee shrubs grown underneath shade trees with feathery leaves. Coffee leaf rust incidence and severity were highest on coffee shrubs growing underneath A. abyssinica (32.3% and 7.5% for incidence and severity, respectively), C. macrostachyus (26.6% and 9.4%) and O. welwitschii (28.4%, 9.0%; Fig. 4f, g). The severity of the hyperparasite was high on shrubs underneath the shade tree species P. fulva, A. abyssinica, and F. sur (8.6%, 7.5% and 6.4%, respectively; Fig. 4i). Coffee berry disease differed significantly among shade tree species, with particularly high infection levels on coffee shrubs growing underneath A. abyssinica and P. fulva (34.2% and 32.9%, respectively; Fig. 4j). While coffee leaf rust, coffee berry disease and hyperparasite incidence were unrelated to canopy cover, hyperparasite severity showed a humpbacked relationship with canopy cover (Fig. S4).
When including soil moisture as a covariate in the pest and disease models, the effect of shade tree species and canopy cover on coffee pests and diseases remained unchanged, suggesting that soil moisture did not mediate the effect of shade tree identity and canopy cover on coffee pests and diseases (Table S5). We detected an independent, negative effect of soil moisture on the leaf skeletonizer, serpentine leaf miner, free-feeding herbivory and antestia, whereas other pests and diseases were unaffected by soil moisture (Table S5, Fig. S5).

Shade tree selection criteria by farmers
We identified 17 criteria used by farmers for selection of desirable shade tree species (Table 1). As the most important criteria, farmers preferred shade trees with feathery leaves (66.1% of farmers), shade Fig. 2. The impact of canopy cover on the incidence of: a) coffee blotch miner, b) serpentine leaf miner, c) coffee leaf skeletonizer, d) free-feeding herbivory, e) presence of antestia, f) coffee leaf rust incidence, g) coffee leaf rust severity, h) hyperparasite incidence, and i) hyperparasite severity on Arabica coffee in a plantation in southwestern Ethiopia. The black circles represent the percentage of infested leaves at the shrub-level, except for panel e, where circles represent the presenceabsence of Antestia at the tree level. The blue solid and dashed trend lines represent significant and non-significant relationships, respectively, as obtained from linear regression models (panels a-d, and f-i) and a generalized linear model with binomial distribution (panel e). The statistical summaries from the models are presented in Table 2.
trees that do not compete for soil moisture with the coffee shrub (57.6%), shade trees that enhance bean quality (49.2%) and quantity (32.2%) and shade trees that retain their leaves during the dry season (30.5%). The impact of shade tree species identity on pests and diseases was ranked 9 th and considered by one out of four farmers ( Table 1). The five most preferred shade tree species were Albizia spp. (96.6% of farmers), Acacia abyssinica (76.3%), Millettia ferruginea (66.1%), Cordia africana (30.5%) and Sesbania sesban (18.65%; Fig. S1). While this matches quite well with those trees perceived by the farmers to have a lower disease incidence (A. abyssinica, A. gummifera, A. schimperiana, and M. ferruginea) it seems that the reasons for these preferences were, in general, because of their feathery leaves and wide crown (which allows the passage of an optimum amount of light to reach the coffee shrubs), as followed by the perceived contribution to coffee yield and quality and soil fertility, and was only to a minor extent based on the criterion of pests and diseases (Table 1). As an example of a species that farmers did not like, most farmers believed that Sapium ellipticum is an inappropriate shade tree for coffee cultivation since it serves as a host for herbivorous larvae. While these larvae do not attack coffee, they attract olive baboons, which often damage the coffee branches while feeding on the larvae. In addition, one farm manager and one smallholder farmer mentioned that the coffee shrubs grown underneath C. macrostachyus had high infestation with coffee leaf rust.

Discussion
We found that shade tree species identity had a weak and inconsistent effect on insect pest damage, whereas canopy cover clearly affected insect pest damage. In contrast to the insect pests, coffee leaf rust and its hyperparasite, as well as coffee berry disease, varied strongly among shade tree species. Pest and disease levels did not differ between coffee shrubs underneath trees with different characteristics (nitrogen-fixing ability, leaf type, crown shape), with the exception for a lower coffee leaf rust severity underneath shade trees with feathery leaves in the smallholder survey. Soil moisture negatively affected the coffee leaf skeletonizer, serpentine leaf miner, antestia and herbivory. Interviews revealed that smallholder farmers consider many criteria for selecting shade trees, such as leaf traits, competition for soil moisture and impacts of trees on coffee quality, and selection criteria related to the impact of shade tree species identity on pests and diseases had only a low priority. Taken together, these findings indicate that there is an untapped potential for targeted selection of shade tree species to reduce disease levels, and, to a lesser extent, pest levels. Our findings also highlight the potential of using tree identity for the sustainable management of pests and diseases, with implications for global agroforestry systems.

The relationship between tree species identity and pests and diseases
Overall, fungal pathogens and their hyperparasites were more strongly and consistently affected by shade tree species identity than insect pests. As no previous study simultaneously focused on pests and pathogens, and most studies focused on single pest species, the generality of the conclusion of a stronger effect of shade tree species identity on pathogens than insects remains unclear. Still, two of the four insect pests were affected by shade tree species identity in one of the surveys, illustrating that the selection of shade tree species can be used as an effective management tool for some species within certain environmental contexts. Such species-dependent findings are reported by other researchers (Room and Smith, 1975;Schroth et al., 2000;Staver et al., 2001). For example, Johnson (2000) found no effect of shade tree species identity (Pseudalbizia berteroana and Inga vera) on potential coffee pests in Jamaica, whereas Bukomeko et al. (2018) reported from Uganda that coffee grown under the shade tree Albizia chinensis had a higher incidence of black coffee twig borer Xylosandrus compactus than coffee grown under Carica papaya, since the sap exudate from the papaya trees repelled X. compactus. In contrast to the insect pests, the incidence and severity of diseases consistently differed among the shade tree species. Both in the plantation study and landscape survey, we found a high incidence and severity of coffee leaf rust on coffee shrubs grown under A. abyssinica and C. macrostachyus trees and a higher severity of its hyperparasite under A. abyssinica, whereas coffee berry disease was higher underneath A. abyssinica and P. fulva.
Several mechanisms might explain the effect of shade tree species identity on the fungal diseases, and, to a lesser extent, the pest species. In this study, we focused on four mechanisms that are often proposed to mediate the effect of shade tree species identity on pest and disease dynamics, namely nitrogen-fixing, soil moisture, leaf type and crown shape (Beer et al., 1998;Gagliardi et al., 2021;Schroth et al., 2000). Nitrogen-fixing bacteria are known to affect crop health by improving soil fertility, which in turn enhances the tolerance of crops for pests and diseases (Beer et al., 1998;Schroth et al., 2000). Yet, we found that pest and disease levels did not differ between nitrogen-fixing and non-nitrogen-fixing shade tree species. In line with our findings, Somarriba and Beer (2011) reported that levels of frosty (Moniliophthora roreri) and black pod rot (Phytophthora palmivora) did not differ among cocoa grown under nitrogen-fixing and non-nitrogen fixing shade trees. While speculative, this suggests that other functional traits of shade tree species are more important in driving pest and disease incidence than nitrogen fixing ability of shade trees (Gagliardi et al., 2021;Sauvadet et al., 2020;Schroth et al., 2000). Shade tree species identity might also affect soil moisture, either through water use by the shade tree, or trait differences such as leaf morphology, foliage density and canopy architecture (Siriri et al., 2013;Tscharntke et al., 2011). While soil moisture explained some of the variation in pest incidence, we did not find any Table 2 The impact of shade tree species identity, canopy cover and the quadratic term of canopy cover on insect pests, as well as coffee leaf rust and its hyperparasite, on Arabica coffee in a plantation in southwestern Ethiopia (see Fig. 1). Data were analyzed at the shrub level (N = 240). Shown are χ 2 -values and p-values from generalized linear mixed effects models. See Table S2  mediating effect of soil moisture on either pests or diseases. The independent, negative effect of soil moisture on the coffee leaf skeletonizer, serpentine leaf miner, antestia and free-feeding herbivory might be explained by the fact that high soil moisture levels promote healthy and vigorous growth, which in turn increases the resistance of coffee plants to insect attack (Schroth et al., 2000;Staver et al., 2001). Shade tree species might also differ in their effect on coffee pests and diseases due to variation in leaf traits and canopy architecture, which can affect important characteristics of the environment such as light interception, air and soil temperature, and rainfall interception Babin et al., 2010;Gagliardi et al., 2021;Gidoin et al., 2014). Yet, we found no consistent imprint of leaf type or canopy shape on pest and disease levels, with only a reduction of coffee leaf rust severity underneath shade trees with feathery leaves in the landscape survey. As severity but not incidence was affected by leaf type, our results suggest that leaf type did not influence the dispersal of H. vastatrix uredospores among coffee shrubs and leaves, but rather affected spore germination and development of uredospores on individual leaves Boudrot et al., 2016). Our finding of lower rust severity on coffee shrubs grown underneath shade tree species with feathery leaves matches with Gagliardi et al. (2021), who reported that the infestation level of coffee leaf rust was lower on coffee shrubs grown under a shade tree with smaller leaves. As a future direction, it will be important to explore the full set of potential mechanisms behind the effects of shade tree species identity on insect pests and diseases with a range of different life-history traits. It might be particularly promising to focus on how leaf and root traits, as well as canopy architecture, affect the microclimatic conditions below the tree canopy, such as light levels, soil temperature, air temperature, soil moisture, air humidity, and rainfall interception (Gagliardi et al., 2021;Staver et al., 2001), and how this in turn affects the dynamics of pests and diseases with different life-histories. In future studies targeting multiple pests and pathogens it would also be interesting to consider the potential importance of intra-and interkingdom interactions between pests and pathogens, as well as management and other local environmental variables, that may affect pest and disease dynamics. Another promising direction is to focus on the impact of the spatial distribution of shade tree species on pest and disease dynamics (Babin et al., 2010;Durand-Bessart et al., 2020;Gidoin et al., 2014).

Impact of canopy cover on insect pests and diseases
Canopy cover had a strong impact on the composition of the pest and Fig. 3. The impact of shade tree species identity on the incidence and severity of coffee leaf rust and its hyperparasite on Arabica coffee in a coffee plantation in southwestern Ethiopia. Stars indicate significance of the model (*** p < 0.001, ** p < 0.01), and letters indicate pairwise differences between shade tree species. Box plots are based on the percentage of infested leaves at shrub-level (N = 60 for each shade tree species). The hatched boxes represent nitrogen-fixing trees. The statistical summaries from the models are presented in Table 2.  Table 3.
disease community by influencing species in different ways, illustrating that the effect of canopy cover on coffee pests and diseases is highly dependent on species-specific life-history traits, and thereby difficult to generalize López-Bravo et al., 2012;Mouen Bedimo et al., 2008;Soto-Pinto et al., 2002). Our observation of a negative relationship between canopy cover and coffee leaf miners matches that of Staver et al. (2001) and Teodoro et al. (2008) on Leucoptera species in Central America, possibly due to lower temperatures and higher humidity under shaded conditions. The decrease of antestia with increasing canopy cover contrasts with Crowe and Gebremedhin Table 3 The impact of shade tree species identity, canopy cover, the quadratic term of canopy cover and management intensity on insect pests, coffee leaf rust and its hyperparasite, and coffee berry disease on Arabica coffee across 58 sites in southwestern Ethiopia (see Fig. 1). Data were analyzed at the shrub level (N = 307). Shown are χ 2 -values and p-values from generalized linear mixed effects models. See Table S3 Table 3.
(1984) and Mugo et al. (2013), who reported higher infestation by antestia in dense shaded coffee plantations. These contrasting observations could be due to differences in methodological approaches between studies. For instance, Mugo et al. (2013) only compared coffee grown in open sun and shaded conditions without quantifying the level of shade, whereas we examined changes in pest levels across a gradient of canopy cover from c. 10-90%. The positive relationship between canopy cover and the coffee leaf skeletonizer corresponds with Samnegård et al. (2014), who found higher damage by the coffee leaf skeletonizer when coffee was growing in a dense contiguous forest than when growing in less-shaded forest patches in southwestern Ethiopia. Our study showed that canopy cover had a weak and inconsistent effect on the incidence and severity of coffee leaf rust and the hyperparasite. For example, in the landscape survey we found that the incidence and severity of coffee leaf rust, as well as the incidence of the hyperparasite and coffee berry disease, were unaffected by canopy cover. Our finding of no effect of canopy cover on the incidence of coffee berry disease matches previous findings in Malawi (Phiri et al., 2001) and southwestern Ethiopia (Zewdie et al., 2020). In contrast to the landscape survey, rust incidence showed a humpbacked relationship with canopy cover in the plantation study, and rust severity showed a negative linear relationship. Our findings of context-dependent effects of canopy cover on rust incidence and severity are reflected in the literature. For example, López-Bravo et al. (2012) reported an increase in coffee leaf rust incidence with increasing shade cover while others have reported the opposite (Soto--Pinto et al., 2002;Zewdie et al., 2021). One promising avenue for further research might be to study the joint influence of canopy cover and shade tree characteristics and microclimatic conditions on the dispersal and development of H. vastatrix spores Gagliardi et al., 2021Gagliardi et al., , 2020. A humpbacked relationship of canopy cover and coffee leaf rust incidence (in the plantation study) and severity of the hyperparasite (in the landscape survey) have not been described previously, but the existence of optimum shade levels and context-dependency highlights the need to compare a range of different shade levels and test patterns in different locations within experimental trials. Overall, the strong relationship between canopy cover and insect pests, and to a lesser degree diseases, supports the notion that the manipulation of canopy cover can be an important management tool (Jonsson et al., 2015;Mouen Bedimo et al., 2007;Staver et al., 2001;Teodoro et al., 2008). At the same time, when taking such an approach it is important to take a holistic perspective, as manipulating the shade to reduce one pest and disease species may increase pest and disease levels of other species.

Shade tree selection criteria by farmers
Farmers' selection of shade tree species is often based on traditional knowledge (Albertin and Nair, 2004;Soto-Pinto et al., 2007;Wagner et al., 2019). Our interviews revealed that farmers consider a wide range of selection criteria. The most important criteria were feathery leaves, competition for soil moisture, impact on coffee quality and quantity, and the retention of leaves during the dry season. Feathery leaves were preferred by farmers since it obstructs strong light, provides a light shade, and allows the passage of rain water to the coffee shrubs. In addition, the small leaflets easily fall from the trees to the ground without damaging the coffee flowers and fruits. Overall, similar selection criteria were reported in some Latin American countries (Soto-Pinto et al., 2007;Valencia et al., 2015), Kenya (Lamond et al., 2019) and Ethiopia (Ango et al., 2014;Hundera, 2016). The farmers top five shade tree species were Albizia spp. (96.6% of farmers), Acacia abyssinica (76.3%), Millettia ferruginea (66.1%), Cordia africana (30.5%) and Sesbania sesban (18.7%). Among these five shade tree species, all but C. africana are leguminous and have feathery leaves with a widely spread crown, and this preference can thus be easily explained by their main selection criteria. The non-nitrogen-fixing Cordia africana has particularly valuable wood, which probably explains the inclusion in this list. Although farmers generally agree on the set of preferred trees species, smallholder farms are not necessarily dominated by preferred trees. This may due to farmers deliberately maintaining some shade tree species for values other than shade (timber or flower for honeybees) and due to policies, that prohibit removing the existing native tree species (Cheng et al., 1998), and the fact that the replacement process will take a long time (Geeraert et al., 2019b). It is noteworthy that some shade tree species are apparently non-preferred, but used very much in both plantations and smallholder sites, such as Croton macrostachyus, because of its fast-growing habit and ability to quickly fill gaps in the canopy (Aerts et al., 2011). While mentioned by a quarter of the farmers, the impact of shade tree species identity on coffee pests and diseases only came on a shared eight to tenth place in the ranking. While there was some overlap between those trees preferred by the farmers and those reported by the farmers to have low pest and disease levels (A. abyssinica, A. gummifera, A. schimperiana, and M. ferruginea), those tree species were mostly selected already based on higher-ranking criteria, such as leaf type. Other studies from Mexico and Costa Rica showed that farmers select shade trees based on their effectiveness in regulating pest and diseases (Cerdán et al., 2012;Segura et al., 2004;Valencia et al., 2015). There might be multiple explanations for the discrepancy between our identification of an important effect of shade tree species on fungal diseases in the plantation and landscape surveys, and the low priority that farmers give to this aspect when selecting shade trees. One reason could be that the relationships between shade tree species identity and diseases are difficult to observe by smallholder farmers, as their plots are structurally diverse and have a diverse assemblage of tree species, and the strength and direction of the effect differs among pests and pathogens. The patterns uncovered might also be hard to observe due to their context-dependency, as indicated by the fact that the patterns are only partly consistent between the plantation study and the smallholder survey. Such context-dependency should also caution against the translation of findings from experimental research stations into advice for smallholder farmers without proper validation trials.

Conclusion
Our findings highlight that different shade tree species and shade levels may affect the levels of pests and diseases differently, and that a consideration of tree species identity and canopy cover is important to design ecologically-informed management of insect pests and diseases of coffee. From a global perspective, we hope that our study stimulates similar studies in a range of agroforestry systems (e.g. coffee, cacao), with the ultimate aim of sustainable agroforestry. From a national, applied management perspective, the current study has relevance for sustainable pest and disease management for the millions of smallholder farmers within the study region.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data Availability
Data will be deposited in the Dryad Digital Repository.